About Sklo

Inspiration & Origins

We were working on an AI observability tool at a cafe in our hometown of Mysore, when an Austrian tourist walked up to us, asked about what we're working on and then asked if we could use AI to make chatbots for his business. We realized that building a product that makes chatbots would be a great starting point for our observability tool as well.

Development Process

Once we started on the chatbot, we became regulars at this cafe, working with Antigravity and Gemini 3 Pro for the frontend.

On the day of the submission, someone I knew from my uni happened to visit the cafe with their family. It turns out their dad is in the real estate business in the USA and they'd like our chatbots as well! Now we're adding the features they'd find useful.

Gemini & Antigravity have massively helped us in speeding up our product development process, allowing use to effectively take on product-manager roles at Sklo instead of spending the majority of our time on writing code.

How it Works

Users create an account, an organization on the dashboard (used for team collaboration within their org) and a chatbot. They then add facts, rules, images (optional) and describe the Chatbot's personality. That's it! The custom chatbot is now ready to be integrated into their website!

Sklo uses Gemini 3 Flash for the chatbot, and will also use it to describe uploaded images. These descriptions will be used for RAG (Retrieval-Augmented Generation) by the vector database to show appropriate images to users during conversations, as we don't want to show every single image to the AI model in production as that would add costs.

The architecture is built for speed and observability:

• Frontend: React + Vite + TailwindCSS
• Backend: FastAPI + Pydantic
• Real-time: WebSockets for live conversation monitoring and config syncing

Challenges Faced

One of the biggest challenges was ensuring real-time collaborative editing of the chatbot's facts/personality/rules across different admin sessions. We implemented a custom WebSocket connection manager to broadcast state updates whenever a fact, constraint, or image is modified.

Future Plans

We are currently refining the observer dashboard to include full markdown support and better session lifecycle management to ensure that every chatbot interaction is perfectly captured and observable. We will spend more time talking to our current and potential future users, understand their needs and iterate based on feedback and then transition to sales/customer acquisition.

Built With

• chromadb
• docker
• gemini
• kubernetes
• python
• react
• rest
• sql
• sqlalchemy
• sqlite
• typescript
• vite